RoSE: Robust, Secure & Efficient Wide-Area Routing

Considering the critical role that the Internet plays in our day-to-day lives, the current routing architecture is surprisingly fragile. Major fiber cuts due to accidents have led to wide-spread loss of Internet connectivity. While these accidents are rare, the majority of the routing problems are due to software bugs, misconfigurations, or human error, and can be prevented. The ultimate goal this project is to improve the fail-over time, service availability, stability, and security of Internet routing infrastructure without compromising its scalability and manageability.

Our research approach hinges upon the observation that modeling individual network components in isolation is not sufficient. This project strives to understand the end-to-end behavior of networks (in terms of performance, reachability, or security properties) by building a model of the dynamic interactions across protocol layers (e.g., IP-routing and application-layer) and between network components (e.g., sets of routers with different policy configurations, packet filters, and firewalls). We believe that the key to support truly ubiquitous computing over heterogeneous networks is to find a reliable route (or routes) that deliver predictable performance.

The research plan for the RoSE project consists of three phases:

Phase 1: Modeling routing dynamics and their implications
The first phase of the project focuses on gaining a thorough understanding the routing dynamics and failure characteristics within currently deployed, large-scale, operational wide-area IP networks. The lessons learned are useful for designing future networks. We gather packet-level measurements and routing information from a Tier-1 ISP's backbone and public peering points. These data are analyzed to address the following questions:

Based on the network measurements and initial analysis, we revisit the definition of "service availability" for IP networks, which should account for instantaneous performance characteristics and routing dynamics. This is analogious to the 99.99 availability measure of the traditional telephone networks. Such measure is crucial in determining whether a specific type of application-level performance requirements can be met, and in determining a meaningful Service-Level Agreements (SLAs). We also explore how our routing failure models impact the design of routers, failure restoration schemes, and other traffic engineering practices.

Phase 2: Multi-layer (or -entity) information sharing for better performance and stability
The second phase of this study will focus on modeling the interactions across multiple protocol layers, e.g., between overlay and underlying IP networks, to identify a set of design principles that ensure their synergistic co-existence.

We will explore the design of a Routing Introspection and Feedback System (RIFS) that will: (1) Provide active feedback from the IP-routing layer to high-level overlay networks and applications for joint optimization, (2) Detect, report and resolve routing anomalies, and (3) Verify the correctness of routing protocols, policies, and router configurations.

Phase 3: Rethinking the design of the Internet
The lessons learned in Phase 1 and 2 will form the basis from which we can revisit the fundamental properties of the current Internet and define a set of design principles for next-generation global Internet. We have begun to explore the feasiblity of creating an overlay control layer for inter-domain policy negotiations, fault-tolerance, and traffic engineering. We will continue to investigate the potential use of overlay networks to improve the stability and security of the underlying IP networks.

People

Faculty
  • C-N. Chuah
    		•  Graduate Students
  • R. Keralapura, ECE
  • J. Mai, ECE
  • S. Raza, CS

    • Collaborators
  • G. Iannaconne, Intel Research Cambridge
  • N. Taft, Intel Research Berkeley
  • S. Nelakuditi, Univ. South Carolina
    		• Past Collaborators
  • S. Agarwal, Microsoft Research
  • S. Bhattacharyya, Sprint ATL
  • C. Diot, Thompsons Labs
  • Z. Zhang, Univ of Minnesota

    • Publications

    Talks

    Technical Reports

    Acknowlegements

    [ RUBINET Home Page]